SMRTR AI

AI stories

Get hand-picked daily summaries of the best, most informative AI articles from around the web. Browse recent summaries, then jump to the original source when you want the full article.

Latest story• Nov 9, 2025

A new Chinese AI model claims to outperform GPT-5 and Sonnet 4.5 - and it's free

Chinese AI lab Moonshot released its new Kimi K2 Thinking model, claiming it outperforms OpenAI's GPT-5 and Anthropic's Claude Sonnet 4.5 on key benchmarks while costing just $4.6 million to train. The open-source model is completely free to access and use, challenging the subscription-based business model that has dominated the AI industry. This development raises questions about whether businesses need expensive...

Read summary Original source

Latest stories

This puzzle shows just how far LLMs have progressed in little over a year

A geometric puzzle that stumped GPT-4o for two hours last June now takes the latest AI model just five seconds to crack. The challenge involves a grid of dots where each node sits exactly one unit away from its neighbors. What...

Read summary Original

Looking Back: The First 100 Days of Our Website’s AI Chatbot

A software company deployed an AI chatbot trained on their website content to understand visitor concerns and generate leads, receiving ~1,900 messages from ~170 visitors over 100 days. While the chatbot effectively revealed...

Read summary Original

Sam Altman Says That in a Few Years, a Whole Company Could Be Run by AI, Including the CEO

OpenAI CEO Sam Altman predicts entire companies, including his own, will be run by AI within a few years, with billion-dollar firms operated by just two or three people alongside AI systems. He believes AI will surpass human CEO...

Read summary Original

Hierarchical Reranking Unleashed: A Step-by-Step Implementation of Enhanced RAG

A new hierarchical reranking system for Retrieval-Augmented Generation (RAG) combines internal knowledge bases with external web searches to dramatically reduce AI hallucinations and improve answer accuracy. This two-stage...

Read summary Original

Freepik CEO - No, AI hasn't killed the stock image market

Freepik CEO Joaquín Cuenca says AI hasn't killed stock photos, as users still prefer traditional assets for authentic content like travel photography. The company now offers both AI tools and original stock images.

Read summary Original

SMRTR AI

Get the next batch of curated stories in your inbox.

This archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.

More recent stories

Building an Interactive AI Agent for Lightning-Fast Machine Learning Tasks

Data scientists traditionally struggle with time-consuming dataset preparation and complex ML workflows that rely on slow CPU processing. NVIDIA developed an interactive AI agent that uses natural language conversations to...

Read summary Original

Three ways Google scientists use AI to better understand nature

Google scientists created three AI tools to protect ecosystems: a satellite deforestation predictor, a species mapping system, and Perch 2.0 for identifying animal sounds to help conservation efforts.

Read summary Original

Hidden Markov Models Explained

Hidden Markov Models use probability to find hidden patterns in sequential data by modeling invisible states that create observable outputs. They solve key problems using algorithms like Viterbi and Baum-Welch.

Read summary Original

Watch Out For Self-Improving Superintelligence On The Horizon

Experts warn that AI systems are rapidly approaching "self-improving superintelligence," where artificial intelligence can autonomously enhance itself without human oversight, potentially spiraling beyond our control. Google's VP...

Read summary Original

Everyone Misses GPT-4o: Why People Prefer it to GPT-5

GPT-5's launch disappointed users despite promises of PhD-level capabilities and fewer hallucinations, with routing system failures making it slower than GPT-4o and coding tasks running four to seven times slower. Users prefer...

Read summary Original

How To Run an Open-Source LLM on Your Personal Computer

Running open-source language models like Llama and Mistral on personal computers is now accessible without cloud subscriptions or servers. Using platforms like Ollama, users can download and operate AI models locally through...

Read summary Original

Anthropic Commits To Preserving Retired Models. What Does This Mean?

Anthropic announced it will preserve weights for retired AI models with significant usage, breaking from the typical tech practice of simply replacing old models with new ones. The company cites three key concerns: customer harm...

Read summary Original

During the AI Journey to AGI, There Is Concern That We Will Be Like Frogs in a Slowly Boiling Pot

Experts warn humanity may fall victim to the "boiling frog" theory during AI's gradual progression toward artificial general intelligence (AGI), where incremental advances mask growing existential risks. Like frogs that fail to...

Read summary Original

Open Source AI And Its Role

AI companies must choose between open source systems that reveal model details versus closed source systems that keep information secret. The U.S. trails China in open source development, raising competitiveness concerns.

Read summary Original

Google plans secret AI military outpost on tiny island overrun by crabs

Google plans to build a secret AI military data center on Christmas Island, home to one of nature's most spectacular events where 100 million red crabs migrate annually from forest to ocean. The project faces environmental...

Read summary Original

MemMachine: Universal local memory layer for AI Agents

MemMachine is an open-source memory layer that enables AI agents to store and recall data from past sessions, creating personalized, context-aware assistants. The system supports multiple memory types and provides...

Read summary Original

I Built My Own AI Agent using n8n — And You Can Too

A programmer recently discovered they could build their own AI agent in an afternoon without writing much code, using nothing more than a visual workflow tool called n8n and some creative thinking. The homemade digital...

Read summary Original

AI Helps Make Transplantable Brains

Scientists at Johns Hopkins are developing lab-grown brain organoids using stem cells converted from human hair, skin, and blood cells to help patients with traumatic brain injuries. These mini-brains include functioning blood...

Read summary Original

Opal: Google’s No-Code AI App Builder Is Now Global

Google expanded Opal, its no-code AI app builder, to over 160 countries. Users can create complete AI applications using natural language instructions without coding, APIs, or infrastructure setup.

Read summary Original

LLMs Encode How Difficult Problems Are

Large language models internally encode problem difficulty that closely matches human judgment, with researchers finding this representation can be decoded with 88% correlation and strengthens during reinforcement learning...

Read summary Original

HeyGen Rides The AI Video Boom Rocket

HeyGen, an AI video startup, reached $100 million in recurring revenue by helping 200,000 customers create marketing videos using AI avatars instead of traditional filming.

Read summary Original

How AI is fueling an existential crisis in education

Generative AI is creating deep philosophical questions in education beyond just student cheating concerns. Teachers are grappling with existential doubts about their purpose as AI increasingly handles course creation, grading,...

Read summary Original

The Rise of the Specialist: Why Small Language Models are the Future of Enterprise AI

The AI industry is shifting from "bigger is better" to specialized Small Language Models (SLMs) that outperform massive systems on specific enterprise tasks while costing 10-30x less to operate. Companies like Microsoft are...

Read summary Original

Back to all archives Load more stories